Software is currently a key part of many safety-critical applications. But the main problem facing the computer industry is how to develop a software with (ultra) high reliability on time, and assure the quality of software. In the past, some researchers reported that the Pareto distribution (PD) and the Weibull distribution (WD) models can be used for software reliability estimation and fault distribution modeling. In this paper we propose a modified PD model to predict and assess the software fault distribution. That is, we suggest using a special form of the Generalized Pareto distribution (GPD) model, named the bounded Generalized Pareto distribution (BGPD) model. We will show that the BGPD model eliminates several modeling issues that arise in the PD model, and perform detailed comparisons based on real software fault data. Experimental result shows that the proposed BGPD model presents very high fitness to the actual fault data. In the end, we conclude that the distribution of faults in a large software system can be well described by the Pareto principle.